P065: Cell-free testing reliably detects DPYD and UGT1A1 PGx alterations, improving chemotherapy dosing and clinical yield*

Methods: In this study, using the national database of the C-CAT, which accumulates real-world data on CGP testing in Japan, we analyzed 169,370 variants detected in 11,399 patients who underwent with FoundationOne Liquid CDx (F1L) testing as a discovery cohort.We then analyzed 72,701 variants detected in 4,763 patients and 9,987 in 2,345 patients who underwent F1L testing and Guardant360 testing, respectively, as validation cohorts.To determine the VAF threshold for discrimination, the VAF distributions of representative predominantly presumed germline variants (rPPGVs)-common polymorphisms with >1% allele frequency in the Japanese population-were compared with representative predominantly presumed somatic variants (rPPSVs)-oncogenic hotspot variants in genes like KRAS and PIK3CA.The detection status of other variants was also analyzed to improve specificity.Additionally, criteria were developed to improve sensitivity by considering information other than VAFs, such as the detection of long indels or reversion variants.Results: The VAF analysis comparing rPPSVs and rPPGVs determined that a threshold of 0.372 VAF achieved the highest area under the ROC curve (AUC 0.974) for distinguishing germline variants.However, to enhance clinical sensitivity based on previous guidelines, a VAF criterion of 0.3 or more was adopted for suspecting germline origin, yielding a sensitivity of 0.988 and specificity of 0.926.A total of 341 rPPSVs (7.4%) exhibited VAFs of 0.3 or higher, leading to the definition of increased somatic fraction (ISF) cases.Finally, the GeMSort algorithm incorporated additional criteria, including a VAF threshold of 0.2 for long insertions/deletions (indels ≥ 10 bp) to account for potential mapping errors that can cause lower VAFs, and a check for somatic reversion variants in BRCA1/2, which can decrease the pathogenic germline VAF.When the GeMSort algorithm was applied to the 26 high-germline conversion rate (GCR) genes, 726 variants were identified as presumed germline pathogenic variants in 710 patients of the discovery cohort.The validation cohort achieved comparable sensitivity and specificity, revealing that the algorithm can consistently differentiate between somatic/germline variants.Conclusion: This study successfully established the GeMSort algorithm for extracting presumed germline pathogenic variants (PGPVs) in LB-CGP testing based on real-world data from over 18,000 patients.This algorithm should help to discriminate with high accuracy whether the variants detected in liquid CGP tests are of somatic/germline origin, although further analyses are required to confirm the validity and utility of this algorithm.